Intelligent Model for Manual Sorting of Plastic Wastes Kudirat Jimoh Anuoluwapo Ajayi

Transcription

International Journal of Computer Applications (0975 – 8887)
Volume 101– No.7, September 2014
Intelligent Model for Manual Sorting of Plastic Wastes
Kudirat Jimoh
Anuoluwapo Ajayi
Oluwatobi Ayilara
Computer Sc. & Eng. Dept.,
Obafemi Awolowo University,
Ile-Ife, Nigeria
Ile-Ife, Nigeria
Ile-Ife, Nigeria
ABSTRACT
The need for an automated sorting system cannot be
overemphasized due to the growing needs for high throughput
and accuracy. Manual sorting method may achieve good
accuracy, but at the expense of throughput. This study
formulates an automated plastic wastes identification model to
overcome the limitation of the manual sorting method. The
proposed identification model employed singleton Sugeno
fuzzy model, image processing and template matching
techniques as its classifier using physical properties of plastics
such as power spectrum of sound, plastic average area and
plastic recycling code as feature set. The model was
developed and simulated in the MATLAB R2012a
environment. The performance evaluation of the model was
carried out using three performance metrics, namely,
accuracy, precision and recall. The model obtained average
accuracy of 0.88, average precision value of 0.77 and average
recall value of 0.25, respectively for the plastic types (PET,
HDPE, LDPE, and PP) used in this study.
General Terms
Plastic identification, Fuzzy logic, Template matching,
spectral analysis, Image processing.
Keywords
Plastic wastes, inference system, sorting, resin code.
1. INTRODUCTION
The amount of plastic waste (sachet water bags, plastic
bottles, wrapping products, and so on) is on the increase in
many countries due to the increase in population. In
developing countries, for instance, the absence of a sound
policy to ensure timely collection, transportation and disposal
of plastic wastes has encouraged people to dispose plastic
wastes on streets with impunity. These wastes find their way
in blocked channels, thereby causing floods and other
environmental hazards. The need for a model to identify
plastic wastes for efficient recycling process is thus desirable.
Sorting of plastic resins is an important step in plastic
recycling and many wide ranges of incompatible plastic resins
abound that may inhibit recycling process. The separation of
different plastics by type is mandatory because contamination
in the recycling of one type of plastic with another can cause
processing problems [10].
Methods for sorting waste plastics range from manual sorting
to advanced automated technologies. The sorting and
separation techniques are often done using chemical, optical,
electrical or physical property differences between various
plastics to be sorted [15]. Density based sorting methods
include float-sink tanks and hydrocyclone. Optical and
spectroscopic-based methods were used to separate plastics on
the basis of colour [13], [19] or type [12], [16-17]. These
techniques cannot identify different grades of the same
polymers and cannot be used when plastic waste surface is
wet. [1] developed an automatic sorting of plastics based on
optical identification of fluorescence signatures of dyes,
incorporated into such materials in trace concentrations. This
work’s limitation is the speed and purity of sorting, which
were limited by the mechanical singulation inadequacy in the
conveyor system at high speeds.
Auto-sort technologies based on X-ray detection [9] was used
to differentiate between PVC and PET bottles as X-ray
sensors was able to detect the presence of chlorine atom found
in PVC bottles. This method cannot identify other plastic resin
types, besides, X-rays are a form of radiation; hence,
precautions must be taken to protect workers from exposure.
In [15], a probabilistic approach was proposed to sort plastic
wastes into PET and non-PET by integrating size, color and
distance. White strips on bottle are first identified by
employing maximum likelihood approach and information on
the white and grey strips is then analyzed by using maximum
a posteriori method.
Other research conducted on the use of artificial neural
networks in conjunction with infrared spectroscopy for
efficient sorting of plastics include [4], [7]. [3] proposed an Xray fluorescence detection system based on the use of tracers
for plastics identification while research based on machine
vision or image processing technique for plastic bottles
identification are discussed in [14], [24].
Truly, automated sorting systems can treat large volume of
plastic efficiently with minimal human intervention but they
often require high investments in specific technologies [6],
[21]. Manual sorting on the other hand, relies on plant
personnel for visual identification and manual extraction of
the plastic containers. However, this is not suitable for plants
with large throughput as efficiency of the process is reduced.
Nevertheless, in many developing countries human labour is
cost effective when compared to investment required for
automated systems, more so manual sorting in recycling
plants is still the preferred procedure [23]. It is therefore
necessary to improve the manual process of sorting polymer
materials so as to make their recycling profitable. For an
economically efficient recycling of plastic wastes, waste
plastics need to be sorted cheaply and automatically into
individual types and grades [5]. This study therefore, proposes
a model, which combined fuzzy logic model, image
processing technique with template matching algorithm to
automate the manual sorting method so as to overcome its
limitations (accuracy and throughput) as well as improve its
efficiency.
The category of plastic materials used in this study are limited
to polyethylene terephthalate (PET), high density
polyethylene (HDPE), low density polyethylene (LDPE) and
polypropylene (PP).
The rest of this paper is organized as follows. Section 2
describes the proposed plastic wastes identification model.
20
Simulation environment and performance evaluation of the
model are discussed in sections 3 and 4. Finally, concluding
remarks are given in section 5.
2. IDENTIFICATION MODEL
Classification problem involves associating an object with
existing classes of objects. Classes may be defined as a set of
objects, or by a set of rules, which define how objects are
classified into given classes. In this study, the classification of
plastic wastes considered only the physical properties so as to
reduce implementation cost. The distinguishing physical
characteristics used for classification purpose are power
spectrum of sound, plastic recycling symbol code and plastic
average area.
Two classification schemes were employed in this study. The
first scheme used template matching to identify plastic types
having the recycling symbol (labeling system instituted by the
Society of the Plastics Industry) inside a triangle printed at the
bottom of the bottle. However, most plastics manufactured
locally are devoid of this symbol and even when present, may
be indistinct. The singleton fuzzy method, which is based on
using fuzzy rules, is employed as a classification scheme for
cases of this nature. Figure 1 depicts a simplified logic of the
proposed plastic wastes identification model. Captured plastic
image and sound clip are input to this model. The model first
tries to locate the recycling symbol on plastic. If present, the
template matching algorithm is used to determine appropriate
plastic type. Otherwise, distinguishing physical characteristics
(spectrum power of the plastic sound and plastic area) are then
computed and fed into the Singleton fuzzy model to determine
plastic type. Key modules of this model are discussed in the
subsections following.
classification. Features such as area, filling fraction of the
boundary box measured in pixel units, and the sound
produced by tapping or rubbing the plastic surface were
recorded. Sound was used as a distinguishing feature since
most plastics produce unique sound to uniquely identify them.
For instance; nylon, a specific type of LDPE resin, produces a
unique sound when compared to other resins when squeezed.
2.2 Template Matching
Template matching is a technique used in classifying an
object by comparing portions of images with another image. It
is the simplest and widely used for detecting targets in image
information [2]. The template matching problem is described
as follows: Assign an unknown target Y into one of C classes,
which are represented by the specified templates {X l} for
l=1,2…,C. The matching process consists of defining a
similarity measure between the objects Y and X l, denoted by
S(Y, Xl) and assigning object Y to class k if S(Y, X k) > S(Y,
Xl) for all l ≠k. This procedure was used to classify resin
codes in plastic wastes using the feature-based approach, an
approach well suited for cases where both reference and
template images have more correspondence with respect to
features and control points.
Plastic resin code samples are first predefined and stored in
the database. Then, the correlation between the input image
blocks and stored resin code samples are computed to identify
plastic type. The advantages of this method is that it is robust
to noise, simple to implement and it takes less time to locate
the candidate resin from input images. Template matching
was accomplished with a 24×24 pixel templates representing
plastic resin codes of plastic waste images (where present).
Matching technique also calculates the error between input
image and stored resin codes templates using the mean
squared error (MSE) metric:
where, template (x, y) and reference (x, y) are the intensity
values of template and input images, respectively. The
templates of all the resin codes will be define with some test
point. Given an image, the resin code to be recognized will be
compare with those templates, and the template with the
highest matched point will be the resin code that appears in
that image.
2.3 Fuzzy Logic Model
Fig 1:Plastic wastes identification model
2.1 Image Processing
The preprocessing operation reduces noise and redundant data
for the purpose of acquiring an optimum-quality image that
will enhance the classifier performance. The preprocessing
operations (image size reduction, image enhancement,
morphology were carried out on the acquired images to
extract necessary features. The main purpose of feature
extraction is to represent the data in a reduced number of
dimensions so as to removing redundancy and improve
The fuzzy inference method used in this system is the TakagiSugeno-Kang (TSK) [18], [20]. Fuzzy rules represent classes
in ‘fuzzy’ terms using linguistic variables. This technique
enables approximate reasoning to be performed. The
technique also improve performance of classification systems
in three ways: through efficient numerical representation of
vague terms and concepts, by increasing their range of
operation in ill-defined environments, and by decreasing their
sensitivity to noisy data. Fuzzy logic-based systems operate
based on the precise and rigorous mathematics of fuzzy sets.
A fundamental concept of fuzzy sets is that an element x may
be a member of set A with varying degrees, i.e., each member
of the set is characterized by its degree of membership within
the set. The degree of membership of element x in set A is
denoted by
. A mapping of the domain interval to its
degree of membership defines the membership function .
The basic structure of a fuzzy inference system consists of the
following: (1) the fuzzifier, which converts the crisp inputs
into fuzzy inputs, (2) a knowledge base, which contains fuzzy
21
rules along with membership functions (3) an inference
mechanism that applies a fuzzy reasoning mechanism to
derive a fuzzy output; and (4) a defuzzifier, which translates
the fuzzy output into a crisp value. The Singleton model,
which is a special case of Mamdani or TSK fuzzy models,
was used due to less computational efforts of its
defuzzification process when compared to Mamdani's fuzzy
inference method [11]. Besides, the Mamdani model lack
accuracy and has high computational cost [22]. The fuzzy rule
for the Singleton model is of the form:
2, signifying the resin type of HPDE. The rules' inferencing
system is shown in the viewer as well.
Ri: If x1 is Ai1 and ……and xn is Ain then y is C
where, C is a fuzzy singleton.
Each linguistic variable is usually associated with a complete
set of membership function that is defined over the entire
operating range of that variable. Figure 2 shows examples of
membership functions set for the two input variables used in
this study - average power spectrum of sound produced from
plastic (“averageSpectrumPower”) and the average plastic
area (“averageShapeArea”). The membership functions are
triangular functions with the widths of membership functions
appropriately determined by the dispersion of the data within
the membership range. The triangular curve is a function of a
vector x, and depends on scalar parameters: a, b and c, as
given by.
(2)
The range of the membership functions for input
"averageSpectrumPower" is from 0 to 3.4e10-5, while the
range for input variable "averageShapeArea" is normalized
with the biggest plastic area size value set as 1 (x-axis) and
the smallest plastic area size set as 0 (x-axis). Subsection 3.4
described how the spectrum power of sound generated from
plastic wastes were computed. Six membership functions
selected for plastic's average spectrum power were 'Zero (Z)',
'Very Low (VL)', 'Low (L)', 'Medium (M)', 'High (H)' and
'Very High (VH)'. The five membership functions selected for
plastic's average area were 'Very Small (VS)', 'Small (S)',
'Medium (M)', 'Big (B)' and 'Very Big (VB)'.
The number of membership functions and the shape of these
functions are an essential part of the knowledge embodied in a
fuzzy logic system. The domain expert usually supplies this
information. This information when combined with the rules
forms a complete knowledge-base for a given application.
After a thorough observation of the data set, the rules were
constructed as showed in Table 1. The minimum aggregation
method is used in the aggregation of the rules in this study to
form the aggregate output of the fuzzy set.
The defuzzification method used in the fuzzy system is
weighted average. The output of the fuzzy singleton model is
the plastic type (PET, HDPE, LDPE, and PP). The fuzzy
system used to identify plastic wastes without resin codes
accepts average power spectrum and average plastic area as
inputs. The output is a single crisp number, which represents
the aggregation and deffuzzication of the inputs. The fuzzy
systems used to process plastic images with their
corresponding sound clips is represented in Figure 4.
Fig 3: Input variables membership functions
Table 1. Fuzzy logic rule set
Ave. Area/
Ave. Spectrum
power
None
S
M
B
VB
Z
-
PP
PP
PP
HDPE
VL
LDPE
LDPE
LDPE
LDPE
LDPE
L
-
PP
PET
PET
PET
M
-
-
PP
PET
-
H
-
PP
PET
PET
HDPE
VH
-
HDPE
PET
PET
HDPE
Figure 5 illustrates the rules viewer given example inputs:
aveSpectrumPower = 1.75e-05dB/Hz and aveShapeArea =
0.423 meters, with the corresponding output as plasticType =
22
Finally, the average power of the sequence xn over the entire
Nyquist interval is computed using
The above expression Pxx(ω) represents the power content of
a signal in an infinitesimal frequency band.
The average power of a signal over a particular frequency
band [ω1, ω2], 0≤ω1≤ω2≤π, is found by integrating the PSD
over that band.
Fig 4: Fuzzy inference system
Periodogram is the simplest nonparametric method used to
estimate PSD directly from the signal itself. In general terms,
one way of estimating the PSD of a process is to simply find
the discrete-time Fourier transform of the samples of the
process (usually done on a grid with an FFT) and
appropriately scale the magnitude squared of the result. This
estimate is called the periodogram. The computation of the
periodogram via an FFT algorithm in some cases, is more
efficient if the number of frequencies is a power of two.
Therefore it is not uncommon to pad the input signal with
zeros to extend its length to a power of two. The periodogram
estimate of the PSD of a sample LDPE sound signal used in
this study, was computed by multiplying its data vector by a
Hamming window. This is as depicted in Figure 6.
Fig 5: Rules viewer with input examples
2.4 Spectrum Power Computation
The spectral analysis of sound produced by a given plastic
waste sample is determined as follows. The average power of
a deterministic periodic signal is computed using the
following: The power spectral density (PSD) of a stationary
random process xn, in terms of normalized frequency, is given
as
Eq (1) when written as a function of a physical frequency f
(using the relation ω = 2πf/fs, where fs is the sampling
frequency) gives
Fig 6: Rules viewer with input examples
3. MODEL SIMULATION
The plastic wastes identification model was simulated in
MATLAB R2012a environment. Sample plastic wastes data
(rescaled to size 90 by 110) with their equivalent wave sounds
(file size between 3KB to 5KB) were processed on an Intel
Core i5 2.50 GHz machine using the model proposed in this
study. MATLAB software was adopted because of its user
friendliness, scalability and many other support features it
provides.
3.1 Data Sample Used
The correlation sequence from the PSD is then derived using
the inverse discrete-time Fourier transform given as:
For the plastic sorting model, image samples of PET, HDPE,
LDPE and PP were collected.
In total, 500 samples from streets and disposal sites were
collected for the test. Sample data including bottles and
containers with different sizes consisted of PET (35%),
HDPE(20%), LDPE(30%) and PP(15%), respectively.
Brief description of the resins used in this study follows:
23
i).
ii).
iii).
iv).
Polyethylene Terephthalate (PET No. 1) containers.
They are commonly used for beverages such as
mineral water and soda beverages. These items are
identified as always being transparent and usually
green or clear in color.
High Density Polyethylene (HDPE No. 2)
containers. They are either rigid or soft. Rigid
HDPE containers are opaque and solid in color,
such as oil containers, household cleaning solution
bottles, base cups of large beverage bottles, etc. Soft
HDPE containers (such as juice and milk bottles)
are semi-transparent and white colored.
Low Density Polyethylene (LDPE No. 4). These are
simply film plastics used for such purposes as
agricultural covering, grocery bags, food industry
wrap, dry cleaning bags, and so on.
Polypropylene (PP No. 5). These are plastics used in
the following products: battery cases, medical
containers, oil additive containers, some dairy tubs,
cereal box liners, bottle labels and caps, rope and
strapping, combs, snack wraps, and bags.
which random sample of plastic images (those with resin
codes and those without resin codes) with their respective
sound signals where applicable, are presented to the model as
input. The classification efficiency of the model using
performance metrics defined above was recorded. Table 2
shows the performance of the model to recognize plastic types
used in this study. The accuracy of the model are 0.88 (PET),
0.86 (HDPE), 0.93 (LDPE) and 0.89(PP), respectively. The
average accuracy (0.89), precision (0.77) and recall (0.25)
values obtained in this model can be attributed to the use of
simple but efficient algorithms, in addition to the feature data
utilized for identification purpose.
Table 2. Performance evaluation of the model
Resin Type
PET
HDPE
LDPE
PP
Average
A
P
R
0.92
0.86
0.93
0.82
0.88
0.73
0.75
0.93
0.66
0.77
0.25
0.28
0.27
0.20
0.25
4. PERFORMANCE EVALUATION
The classification results is shown in graphical form (bar and
pie charts) for easy interpretation. Figure 3 depicts sample
output indicating the model’s accuracy in plastic identification
problem. Fig. 3(a) is a sample input of plastic images
consisting of plastics with recycling code and those without
recycling code. Outputs from the model are presented in two
forms- bar chart and pie chart. The bar chart (Fig. 3b) shows
the classification of each plastic sample into the appropriate
resin type (PET, HDPE, LDPE, and PP), while the pie chart
(Fig. 3c) depicts the frequency of each plastic type. The three
metrics used to evaluate the performance of this model are
accuracy (A), precision (P) and recall (R).
Given that Tn is the total number of sampled plastics, tp is the
true positive detection of a specific plastic type, fp is the false
positive detection of a particular plastic type, tn is the true
negative detection of a given plastic type and fn is the false
negative detection of a given plastic type, then
5. CONCLUSION
In this study, a singular fuzzy logic model, template matching
and image processing techniques were applied to automate the
manual process of sorting and identifying four plastic types
(PET, HDPE, LDPE, and PP). The use of fuzzy logic as an
alternative to other intelligent techniques such as neural
networks and genetic algorithms provides a computational
inexpensive alternative.
The system developed in this study attempts to emulate the
manual process of plastic wastes identification using physical
properties such as resin code, area and sound wave in a more
timely and effective manner. The model was developed and
its performance evaluated. It was observed that the model was
capable of classifying plastic wastes using physical
characteristics such as sound clips, recycling code, and
plastic area of plastics. It is also observed that sound signal
can be used in plastic waste identification.
This model can be used to separate plastic wastes before they
are sent to landfill sites to lower cost implication. Since most
of the current commercial technologies have process
capabilities and price ranges that limit their use to large
volume recyclers, in future, we intend to build a robot that
will incorporate the model developed in this study as its
vision, and used in place of human to optimize plastic wastes
recycling process.
To evaluate the effectiveness of our model to recognize
plastic types used in this study, we made experiments in
24
Fig 5: Classification results (b and c) for the sample plastic images in (a) as input
[8] Hayes, M.H. Statistical Digital Signal Processing and
Modeling. New York: John Wiley & Sons, 1996.
6. REFERENCES
[1] Ahmad, S.R., 2004. A new technology for automatic
identification and sorting of plastics for recycling.
Environ. Technol. 25 (10), 1143–1149.
[2] Bae, J.S and Song, T.L. 2008. Image Tracking Algorithm
using Template Matching and PSNF-m. International
Journal of Control, Automation, and Systems, 6(3), 413423
[3] Bezati, F. Froelich, D. Massardier, V. Maris, E. 2010.
Addition of tracers into the polypropylene in view of
automatic sorting of plastic wastes using X-ray
fluorescence spectrometry, Waste management, 30,591–
596
[4] Coghlan, A. 1993. Infrared Makes Light Work of Sorting
Plastics. New Scientist, 139(1882), 20.
[5] Froelich, D., Maris, E., Haoues, N., Chemineau, L.,
Renard, H., Abraham, F., Lassartesses, R. 2007. State of
the art of plastic sorting and recycling: feedback to
vehicle design. Miner. Eng. 20 (9 SPEC. ISS.), pp. 902–
912.
[6] Hearn, G.L., Ballard, J.R. 2005. The use of electrostatic
techniques for the identification and sorting of waste
packaging materials. Resour. Conserv. Recycl., 44(1):9198
[7] Huth-Fehre, T. Feldhoff, R. Kantimm, T. Quick, L.
Winter, F., Cammann, K., van den Broek, W. Wienke,
D., Melssen, W. and Buydens, L. 1995. NIR-Remote
Sensing and Artificial Neural Networks for Rapid
Identification of Post Consumer Plastics. Journal of
Molecular Structure, 348, 143-146.
[9] Inculet, I. 1992. Electrostatic separation of mixed plastic
waste Proc. Davos 92 Recycle
[10] Masoumi H, Safavi S. M., and Khani Z. 2012.
Identification and Classification of PlasticResins using
Near Infrared ReflectanceSpectroscopy. International
Journal of Mechanical and Industrial Engineering 6,213220
[11] Mclvor, R.T, McCloskey, A.G, Humphreys, P.K,
Maguire, L.P. 2004. Using a fuzzy approach to support
financial analysis in the corporate acquisition process.
Expert Systems with Applications, 27,533-547
[12] Moore, S. 1999. Infrared scanner affords easy plastic
identification. Mod. Plast. 76, 32–33.
[13] Safavi, S.M., Masoumi, H., Mirian, S. S., Tabrizchi, M.
2010. Sorting of polypropylene resins by color in MSW
using visible reflectance spectroscopy. Waste
Management 30, 2216-2222.
[14] Scavino, E., Wahab, D.A. Hussain, A. Basri, H. and
Mustafa. M.M. 2009. Application of automated image
analysis to the identification and extraction of recyclable
plastic bottles. Journal of Zhejiang University of Science
A, 10(6):794-799
[15] Scheirs, J. 1998. Polymer Recycling, Wiley Publication,
[16] Scott, D. M. 1995. A Two-Colour Near-Infrared Sensor
for Sorting Recycled Plastic Waste. Measurement
Science and Technology, 6(2), 156 -159.
[17] Sommer, E. J. 1996. Methods and technologies for postconsumer plastic identification and sorting. Proc. SPE
3rd Annual Recycling Conf., Chicago, IL, pp. 299-304.
25
[18] Sugeno, M. and Kang, G.T. 1988. Structure
identification of fuzzy model, Fuzzy Sets and Systems,
28, 15-33.
[19] Tachwali, Y. Al-Assaf, Y.
Al-Ali, A.R. 2007.
Automatic multistage classification system for plastic
bottles recycling, Resources, Conservation & Recycling.
52(2), 266-285
[20] Takagi, T. and Sugeno, M. 1985. Fuzzy identification of
systems and its applications to modeling and control,
IEEE transactions on Systems, Man, and Cybernetics 15,
116-132.
[21] Taylor, B. 2006. Visions of plastic. Recycl. Today,
44(4), 48-56.
Fuzzy Model using Memetic Algorithm .Information
Technology and Computer Science, 8, 24-37
[23] Wahab, D.A., Abidin, A., Azhari, C.H. 2007. Recycling
trends in the plastics manufacturing and recycling
companies in Malaysia. J. Appl. Sci., 7(7), 1030-1035.
[24] Wahab, D.A., Hussain, A., Scavino, E., Mustafa, M.M.,
Basri, H. 2006. Development of a prototype automated
sorting system for plastic recycling. Am. J. Appl. Sci.,
3(7), 1924-1928.
[25] Zulkifley, M. A, Mustafa, M. M and Hussain, A. 2013.
Probabilistic White Strip Approach to Plastic Bottle
Sorting System,, IEEE International Conference on
Image Processing, Melbourne, Australia. pp: 3162-3166.
[22] Wadhawan, S., Goel G. and Kaushik S. 2013. Data
Driven Fuzzy Modeling for Sugeno and Mamdani Type
IJCATM : www.ijcaonline.org
26

Intelligent Model for Manual Sorting of Plastic Wastes Kudirat Jimoh Anuoluwapo Ajayi

Transcription

Similar documents

"Inteplast Group: Turning Plastic Into Gold" from American Executive

Plastics expertise applied: MuCell∏. Progress for your success

Ad, 8.625x11.125, Laverson, RanchandCoast

Kawasaki 2000

Mini Maracas 3+ months

Shofar Template

Plastic? - Mercator Energy

ist 221 - Royal Model